📈 Machine Learning Pipeline for Stock Prediction

Predicting Next-Day Stock Prices Using 10-K, 10-Q, and 8-K Filings

---

🧠 Overview

This project demonstrates a comprehensive Machine Learning (ML) pipeline to predict next-day stock prices using recently published SEC filings (10-K, 10-Q, and 8-K). The work was conducted as part of my Master Thesis, showcasing end-to-end data processing, model development, evaluation, and a trading simulation.

---

🏗️ Pipeline Architecture

🔬 Data Preprocessing

• Merging and cleaning financial text data (not included, done in another part)
• Handling missing/corrupted values
• Data normalization before training and denormalization for simulation
• Exploratory analysis of:
  • Training/test/validation splits
  • Distribution of 8-K vs. 10-K/10-Q filings
  • Label statistics and dataset health (e.g., NaNs)

📦 Dataset & Dataloader

• Custom Dataset using masked input arrays
• Train/Test/Validation split
• Tensor shape adjustments for compatibility
• Efficient loading using PyTorch's DataLoader

🤖 Model

• Implemented in PyTorch
• Simple LSTM model for time-series analysis
• Support for hyperparameter tuning and configuration

🔁 Training & Testing

• Training and evaluation loops with:
  • Loss tracking
  • Metric logging
• Explanation and impact of key hyperparameters

---

📊 Evaluation Metrics

📉 Regression Metrics

• Mean Absolute Error (MAE)
• Mean Squared Error (MSE)
• Root Mean Squared Error (RMSE)
• R-Squared (R²)

🧮 Classification Metrics

• Accuracy
• Precision / Recall / F1-Score
• Confusion Matrix

📊 Data Exploration

• Sample counts (train/test/val)
• Label count & distribution
• Change in values from today to next-day
• Comparison: 8-K vs. 10-K/Q filings
• NaN statistics and label prediction counts

---

📈 Visualizations

• 📚 Learning Curves
• 🧠 SHAP (SHapley Additive exPlanations) analysis for model interpretability
• 🗂️ Dataset distribution and breakdown charts
• 📉 Prediction vs. Ground Truth plots
• 📊 Confusion matrices

---

💰 Trading Simulation

A simple 'All In' strategy simulation, comparing:

• ✅ Our model's predictions
• 🧍‍♂️ Buy and Hold
• 🎲 Random Walk
• 📈 Always Long
• 📉 Always Short

🔧 Features:

• Customizable commission fees (e.g., $5/side in showcased run)
• Visualization of equity curves
• Tabular summary of simulated performance

---

🛠️ Technologies Used

• Python
• PyTorch
• Pandas / NumPy
• Matplotlib
• SHAP
• Scikit-learn

---

📚 License

This project is part of academic research and open for educational use. Please cite appropriately.

---

🙋‍♂️ Author

Leonard Müller
Part of my Master Thesis - Julius-Maximilians-Universität Würzburg
Contact: leonardmueller.lm@gmail.com